stats_2023

[zenrabbit]

Ideas

• pop you list of Qs and current work here for steph and I to chew on
• check temp as category vs regress based
• do box plot of light and sensor measures
• build glms as quaisipoisson for counts
• get html link rebuilt

[MarioZuliani]

Boxpot for temp complete. Category vs regression figures and stats complete. Quasipoissons complete. Html link should be working.

Qs: What other types of figures and data do we need? Any updates on other work for this?

[zenrabbit]

1. who prefers warmer? (temp trial data)
2. who lives in the warmest spots? (gbif data for region, then maxent sdms on temp for spots where they live)

[zenrabbit]

see issue on my repo with workflow, lets get alex to check

[zenrabbit]

part 2 of paper

Great idea. Exactly, as you sketched out is what I would recommend i.e., 1) scrape GBIF (can use rgbif package in R), 2) scrap climate from WorldClim (there are others, but WordClim Easiest), 3) extracted climate variables, 4) build some model.

If its only 2-3 climate variables, like you have with Tmax & tmean, then you can go one notch simplifier with a linear model instead of MaxEnt.

The data you will pull down will be presence-only so you will need to generate psuedo-absences. There is a lot of language behind what is a psuedo-absence such as in this paper https://esajournals.onlinelibrary.wiley.com/doi/abs/10.1890/07-2153.1 For your purposes though, it would really be just generating random points between some arbitrary distance from your occurrence points. For example, create a 50 km buffer around all the occurrences, sample an equal amount of points as your occurrence dataset, then use those are absences to create a presence/absence dataframe. If you go the simple linear model route, then it would be just a matter of fitting that all into logistic regression.

Then whatever model output is created, you can use it to predict onto the worldclim rasters to get a visual representation of where the species is/isn't for California. This is all essentially what we did in the Oikos paper, so perhaps start there.

I'm not super familiar with your study, but I could imagine something like a histogram of the species actually temperature occurrences, where you have vertical lines to indicate the temperature thresholds you set in your model. That would be informative.

One final idea. If you were able to do a logistic regression using your own trial data (e.g. germination yes/no ~ tmax * tmin), you could use the output of that model to predict across the California the same you would with the GBIF and worldclim model. It would be kinda cool to see how much congruence there is between the two models. The more similar those model outputs are, the more the species is determined by temperature and not other factors (species interactions, geology, precipitation). This idea is potentially too much though for your paper, and just me spit balling.

[zenrabbit]

Feb 2023 review

1. can you tidy all response variables in into one plot with facets, then use colour for species?
2. I would treat temp as a continuous var and not categorical but then do we end up with stacked ones?
3. organize a table of means perhaps or slopes per species per variables, ie 4 species x response variables, then whether temp had an effect? be useful for paper.

[zenrabbit]

• [x] add a section called 'publication results' right at top of html
• [x] put in key figures and stats that have all other stats after, let's see the figures you are proposing for the paper
• [x] call the responses germination, establishment, and mass per pot? in g
• [x] use broom::tidy for your emmeans and make nice little tables please
• [ ] was temp for a given 'temp category' the same or different within that? Steph checked
• [ ] get worldclim for climate niche
• [ ] MAP, MAT all you need I think for climate niche models
• [ ] tidy up gbif using Marina code? needed?
• [ ] if you want to facet x measure, then you will have pivot_longer etc or you can have separate figures for each response